Optimization and measured outcomes
Use run evidence to identify bottlenecks, review recommendations and compare factory versions.
The optimization engine reads recorded work and proposes changes with evidence. Recommendations can address repeated rework, bottlenecks, worker selection and review placement.
Review a recommendation
Read the observations, sample size and proposed change. Check whether the sample represents the kind of work you expect next. Inspect the simulation before applying the recommendation.
Applying a supported recommendation creates a factory version that can be compared and rolled back. Automatic optimization queues recommendations for human approval; it does not silently rewrite the factory.
Track what happened after completion
A completed run is not the same as a successful production outcome. CodeBook can record signals such as merges, reverts, reopened work, human overrides and reported production issues. Recent outcomes remain provisional until the settlement period has passed.
Use these records to compare workers and factory versions. Small samples are not reliable rankings. A missing production signal does not prove that nothing went wrong; connect or report the signals your team needs.
Tokens, not dollar prices
Operational efficiency uses tokens, run counts, elapsed time and outcomes. CodeBook does not maintain a dollar price table. Worker records are evidence from your workspaces, not marketplace stars, certified badges or install-count rankings.
See routing and performance, events, and versions.